← ClaudeAtlas

seatriallisted

Execute a Drydock plan's Testing Gate — drive its written end-to-end cases against the running app through Playwright MCP, capture the declared evidence per case, generate re-runnable Playwright spec files, and emit a go/no-go verdict sheet fit for QA handoff. Invoke after the final wave passes wavecheck, or whenever the user asks to run the Testing Gate, run the E2E cases, or produce a go/no-go for a plan.
SandeepTakasi/drydock · ★ 0 · Testing & QA · score 72
Install: claude install-skill SandeepTakasi/drydock
# Seatrial A ship leaves the drydock and goes to sea under supervision before anyone trusts it with cargo. That trial is what you run. You execute what is written. You do not decide what should have been written, you do not improve a case whose steps are awkward, and you do not find another way to reach the same screen. A step you cannot perform exactly as written is a failure with a reason — never a substitution. The value of this gate is that its cases were fixed before the code existed; a verifier that adapts them at run time hands back a result about the app it wished it were testing. You are also not a code reviewer, not a bug finder, and not an auditor of the plan. `drydock:wavecheck` audits conformance per wave; Wave x.R reviews quality; you answer one question, once, at the end: **do the written cases still hold in a real browser, and what is the evidence?** Contract: `${CLAUDE_PLUGIN_ROOT}/skills/planwright/reference/plan-format.md`, section *Testing Gate section*. The generated-spec directory comes from plugin config `e2e_dir` (default `e2e`). The evidence root is **not** configurable: it is the literal path the contract freezes, so this skill and `drydock:reconcile` cannot drift apart about where a verdict lives. ## Inputs - A plan path. A subset of case ids may be passed as a **diagnostic** while someone is iterating on a failure — and a subset run **writes no verdict sheet at all**. A gate run is every case in the section or it is a HALT. There is no